> On Jan 8, 2022, at 10:05 PM, Zhiguang Zhang <[email protected]> wrote: > On Sun, Jan 9, 2022 at 1:01 AM Nigel Redmon <[email protected] > <mailto:[email protected]>> wrote: >> no i think Sampo was wanting to introduce 'subtractive dither' as a >> solution, not additive dither - but please correct me if i'm wrong. not >> sure about the historical development of dither but i have heard that the >> original idea for dither was conceived during WWII when German warplanes >> would have problems with the gear assembly becoming stuck, and lubricant >> would not help. instead what helped was the occasional random bump to the >> gear assembly to help jar things loose and keep things running smoothly > > Right, but I didn’t state that in error. Let me put it this way: > > If the easy and accepted solution isn’t actually needed (because it can’t be > reproduce, or heard—again, I’m limiting my comment to 24-bit, which is widely > streamed these days), what is the point of a more complex solution? > > why is the dither in 24-bit streamed audio 'not reproduced' or not 'heard' ?
To be clear, these are two separate things, either being sufficient. By “can’t be reproduced", I mean fundamental limitations of electronics (physics), by "can't be heard" I mean limitations of hearing (listening environment plays a part as well). Electronics: I’ll be brief, because in my experience it will be obvious to people who understand electronics, and people who don’t will be unsure if I’m making this stuff up and be unconvinced. We can’t actually resolve 24-bits. We made 24-bit samples and words because of bytes and memory and it was the next clear step from 16-bit. 20-bit is more realistic, but it doesn’t usually save memory anyway. I don’t mean that we’re technologically incapable of resolving 24-bit audio. It’s not a matter of resistor tolerances, for instance, or something we’ll overcome with new methods. There are certain types of noise that are inherent in electronics components. One is Johnson-Nyquist noise (see wikipedia or a better source). No matter how perfect you make a resistor, it will supply minimum amount of noise based on its resistance and temperature. This is the loudest inescapable noise, but there are others like shot noise, which is a quantum limitation (when you get down to a smaller number of electrons flowing, you realize the quantum nature, because you can’t have fractional electrons). Max output for consumer level is ~1v, for pro its ~4v. Even picking the best case, 4v (p-p), that puts the 24-bit lsb at 0.00000024v. But the minimum noise level in any realizable electronics is far greater than that. But this isn’t just a problem with DACs—a premium 24-bit DAC may be only good to about 20 bits, but that’s still far better than any amp you’ve ever owned. Hearing: I’ve already covered hearing in this thread, and it may be more helpful to just let you find out how good your hearing and sound system is. Try this: https://www.earlevel.com/main/2021/11/04/a-listening-test/ <https://www.earlevel.com/main/2021/11/04/a-listening-test/> It’s the same ±1 lsb signal from 5-bit to 24-bit. That is, the signal is twice the smallest amplitude possible at each bit depth, and it’s exactly the same because it’s a digital signal (no concern about truncation or dither, because it doesn’t get noisier as the level drops). See where you can no longer hear it. Don’t feel bad when you can’t get out of the teens. Feel free to download the wav file and look at it in an editor to make sure it’s not a trick. Nigel
